125
1 INTRODUCTION
The safe conduct of a commercial ship in situations of
passing by with more of the ships they meet is to keep
the least risk of collision with each of them. The
source of information on the current state of the vessel
traffic control process and the collision risk
assessment is the Automatic Radar Plotting Aid
(ARPA) radar anti-collision system, which enables the
operator of the vessel through the TRIAL
MANOEUVRE function to determine the anti-
collision manoeuvre in relation to the most dangerous
vessel, including the rules of the International
Regulations for Preventing Collision at Sea
(COLREGs) maritime route according to International
Maritime Organization (IMO). The COLREGs rules
apply only to two ships met, separately for the
conditions of good and restricted visibility at sea.
Over 80% of ship collisions are caused by the
human factor, during the subjective assessment of the
navigational situation and maneuvering decision in
the game environment (Engwerda 2005, Kun 2001,
Lebkowski 2018, Millington & Funge 2009).
It is assumed that about half of these losses can be
avoided using computer-aided navigator decision
support software using artificial intelligence, game
theory and optimization methods (Bist 2000, Isaacs
1965, Lazarowska 2017, Miloh 1974, Perez 2005,
Szlapczynski & Szlapczynska 2016).
The aim of this article is to present a multi-step
matrix game that contain the risk of ship collisions,
determining the safe trajectory of the ship in terms of
multi-criteria optimization, allowing for the degree of
cooperation in avoiding collisions (Kouemou, 2009,
Lisowski 2016, Olsder & Walter 1977).
This article proposes a new mathematical
formulation model of the collision-risk index,
depending on ships proximity parameters and the
distance between them.
Multi-criteria Optimization of Multi-step Matrix Game
in Collision Avoidance of Ships
J
. Lisowski
Gdynia Maritime University, Gdynia, Poland
ABSTRACT: This research article formulates a mathematical model of the matrix game of the safe ship control
process containing: state variables and control, collision risk definition and the form of a collision risk matrix.
Multicriteria optimization of the matrix game was introduced, leading to non-cooperative and cooperative
game control algorithms and non-game control. Simulation safe trajectories of own ship for various types of
control were compared to the example of the real situation at sea.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 13
Number 1
March 2019
DOI: 10.12716/1001.13.01.12
126
2 THE MATHEMATICAL MODEL OF A MATRIX
GAME
Figure 1 presents a matrix game of the process of
controlling your own ship in the situation of passing
by with the ships at sea.
Figure 1. Matrix game of ships.
2.1 State and control variables
Process state variables are represented by the distance
between the encountered ships D
j, bearing Nj and the
risk of collision r
j and the coordinates of the own-ship
position (X,Y) on the reference trajectory of movement
(Kazimierski & Stateczny 2013).
The control variables of the own-ship are: the
course
and the speed V, and the control quantities
of the j-th encountered ship in question are its course
j and the speed Vj (Tomera 2012).
State variables in the ship control process in
collision situations are measurable by means of the
on-board ARPA anti-collision system.
2.2 Collision-risk
The forms of the collision-risk r
j of own-ship with the
j-th encountered ship as the mean square reference of
a safe situation are determined by the value of the
assumed safe distance D
s and safe approach time Ts
with the current situation of the ships approximation
which are determined by the smallest expected
distance D
j,min and time to the largest approximations
of T
j,min and the distance between ships Dj:
r
j
s
w
,
s
j

a
1
D
s
D
j
,min
s
w
,
s
j

æ
æ
æ
2
a
2
T
s
T
j
,min
s
w
,
s
j

æ
æ
æ
2
D
s
D
j
æ
æ
æ
2
(1)
where the coefficients a
1 and a2 determine: the type of
visibility at sea - good or restricted and the type of
navigation area - open or closed (Modarres 2006, Xu
2014) (Fig. 2).
Figure 2. Collision-risk characteristics in the situation of
concentrated ship traffic at: a
1 = a2 = 0.4, Ds/Dj = 0.5 (above)
and in the situation of larger distances between ships at: a
1 =
a
2 = 0.5, Ds/Dj = 0.1 (below).
The mathematical collision-risk model proposed
here was taken into account not only the approach
parameters of D
j,min and Tj,min, but also the distances
between ships
Dj. The value of collision-risk also
depends on the density of the position of the
encountered ships (Lisowski 2014).
2.3 Matrix of collision-risk
The game matrix R, in which own ship with clean s
w
strategies and m encountered ships with clean
strategies, can each be presented each in the following
equation form:
R
r
j
s
w
,
s
j

æ
æ
r
1
s
w
1
,
s
11

r
1
s
w
1
,
s
12

...
r
j
s
w
1
,
s
j
1
r
1
s
w
2
,
s
11

r
1
s
w
2
,
s
12

...
r
j
s
w
2
,
s
j
1

...
r
1
s
wW
,
s
11

r
1
s
wW
,
s
12

...
r
j
s
wW
,
s
j
1

(2)
The numbers of rows are corresponded to the
amount W of the acceptable own-hip's strategy in the
S
w set, in the form of changes in its course  and the
speed
V to avoid collision (Mesterton-Gibbons 2001,
Polak et al. 2016):
127

12 3
0, , , ...
1, 2, ...,
wwi w w w
Ss s s s V
iW

(3)
The numbers of columns were corresponded to the
total amount of K = mW
j admissible strategies for all m
ships in the set S
j, in the form of changes in their
courses
j and speeds Vj in order to avoid
collisions:

12 3
0, , , ...
1, 2, ...,
jji j j jj j
j
Ss s s s V
jW

(4)
The Constraint-set C(s
w,sj) of the permissible
strategies result from the rules of the right of the
COLREGs sea route are shown in Figures 3 and 4.
Figure 3. Met j-th ship on the starboard side: the sets of
acceptable game strategies of own-ship S
w and j-th
encountered ship S
j (above), and hyperplane of collision-risk
r
j depending on the  own-ship course changes and the
encountered j-th ship 
j (below).
Figure 4. Met j-th ship on the port side: the sets of
acceptable game strategies of own-ship S
w and j-th
encountered ship S
j (above), and hyperplane of collision-risk
r
j depending on the  own-ship course changes and the
encountered j-th ship 
j (below).
3 TYPES OF OPTIMIZATION OF THE MULTI-STEP
MATRIX GAME
In most real transport and logistics processes, the
matrix game does not reach the saddle point and then
has no solution when using pure strategy in game
theory. Therefore, an approximate solution to the
game is the use of the mixed strategy chain as the
probability of using pure strategies (Ehrgott 2005,
Ehrgott & Gandibleux 2002).
First, the probability matrix for using pure
strategies is determined as shown in equation 5:
p
j
s
w
,
s
j

æ
æ
p
1
s
w
1
,
s
11

...
p
j
s
w
1
,
s
j
1
p
1
s
w
2
,
s
11

...
p
j
s
w
2
,
s
j
1

...
p
1
s
wW
,
s
11

...
p
j
s
wW
,
s
j
1

(5)
Then, the most probable strategy is the optimal
control used of the own-ship (Eshenauer et al. 1990,
Osborne 2004):
128

,
max
wj
ss
ww j
uup



(6)
The game ends after bringing the collision-risk of
each ship to the value of zero and reaching a certain
value of the final payment p
f, in the form of the final
deviation of the safe trajectory from its initial value.
3.1 The criterion of a non-co-operative matrix game
The algorithm of MMG_nc multi-step non-co-
operative matrix game uses the following
optimization criterion (Nisan et al. 2007, Straffin
2001):

,
min max
wj
w
j
s
s
nc j
s
s
Qp
(7)
3.2 The criterion of the co-operative matrix game
The MMG_c algorithm of multi-step co-operative
matrix game uses the following optimization criterion
(Basar & Bernhard 2008, Wells 2013):

,
min min
wj
wj
s
s
cj
ss
Qp
(8)
3.3 The criterion of non-game control
The MNG algorithm of multi-step non-game control
uses the following optimization criterion:
min
w
w
s
ng j
s
Qp
(9)
The MMG_nc, MMG_c and MNG algorithms for
determining a safe own-ship trajectory in a collision
situation were developed using the linprog function
from the Matlab Optimization Toolbox package
(Breton & Szajowski 2010).
The method of entering the initial data for
calculations describing the navigational situation is
shown in Figure 5, and the form of calculations results
of the safe trajectory of the ship is illustrated in Figure
6.
Figure 5. Algorithm window with initial data of the
navigational situation.
Figure 6. The algorithm window with the results of
calculations of the safe own-ship trajectory and own-ship
control by change of speed V or course .
4 COMPUTER SIMULATION OF GAME
CONTROL ALGORITHMS
The safe trajectories of own ship in the situation of 33
ships in the Kattegat Strait, in conditions of good and
restricted visibility of the sea, are determined
according to algorithms of multi-criteria optimization:
129
MMG_nc, MMG_c and MNG, and this are shown in
Figures 7-13.
Figure 7. The six-minute speed vectors of own ship and 33
encountered ships in a navigational situation in Kattegat
Strait.
Figure 8. A safe trajectory of own-ship in non-co-operative
matrix game, in conditions of good visibility in the sea, for
D
s = 1.0 nm, final payment pf = 1.88 nm (nautical mile).
Figure 9. A safe trajectory of own-ship in non-co-operative
matrix game, in conditions of restricted visibility of the sea,
for D
s = 2.6 nm, pf = 2.18 nm (nautical mile).
Figure 10. A safe trajectory of own-ship in co-
operative matrix game, in conditions of good visibility
in the sea, for D
s = 1.0 nm, pf = 0.57 nm (nautical mile).
130
Figure 11. A safe trajectory of own-ship in co-operative
matrix game, in conditions of restricted visibility of the sea,
for Ds = 2.6 nm, pf = 1.19 nm (nautical mile).
Figure 12. A safe trajectory of own-ship in non-game
control, in conditions of good visibility on the sea, for D
s =
2.6 nm, p
f = 0.69 nm (nautical mile).
Figure 13. A safe trajectory of own-ship in non-game
control, in conditions of restricted visibility on the sea, for
D
s = 2.6 nm, pf = 0.71 nm (nautical mile).
5 CONCLUSIONS
The use of a matrix game with collision-risk for the
synthesis of algorithms for computer-aided
navigating maneuvering decision makes it possible to
take into account the degree of indeterminacy of the
navigational situation caused by the imperfection of
the law of the sea route and the subjectivity of the
navigator making the maneuvering decision to avoid
a collision.
The multi-criteria approach to the task of
optimizing the safe control of the ship's movement
allows taking into account of both the non-co-
operative and co-operative game control and the
control of the non-target ships.
Obtained safe ship trajectories differ mainly in the
value of the final deviation from the set trajectory of
the movement.
REFERENCES
Basar, T. & Bernhard, P. 2008. H-Infinity optimal control
and related mini-max design problems: A dynamic
game approach. Berlin: Springer.
Bist, D.S. 2000. Safety and security at sea. Oxford-New
Delhi: Butter Heinemann.
Breton, M., & Szajowski, K. 2010. Advances in dynamic
games: theory, applications, and numerical methods for
differential and stochastic games. Boston: Birkhauser.
Ehrgott, M. 2005. Multicriterial optimization. Berlin:
Springer.
Ehrgott, M. & Gandibleux, X. 2002. Multiple criteria
optimization: state of the art annotated bibliographic
surveys. New York: Kluwer Academic Press.
Engwerda, J.C. 2005. LQ dynamic optimization and
differential games. New York: John Wiley & Sons.
131
Eshenauer, H., Koski, J., Osyczka, A. 1990. Multicriteria
design optimization: procedures and application. Berlin:
Springer-Verlag.
Isaacs, R. 1965. Differential games. New York: John Wiley &
Sons.
Kazimierski, W. & Stateczny. A. 2013. Fusion of Data from
AIS and Tracking Radar for the Needs of ECDIS, Proc. of
the Signal Processing Symp., Jachranka, Poland.
Kouemou, G. 2009. Radar technology. Chapter 4 by Józef
Lisowski: Sensitivity of safe game ship control on base
information from ARPA radar. Croatia, In-tech, 61-86.
Kun, G. 2001. Stabilizability, controllability, and optimal
strategies of linear and nonlinear dynamical games.
PhD. Thesis. Aachen: RWTH.
Lazarowska, A. 2017. A new deterministic approach in a
decision support system for ship’s trajectory planning.
Expert Systems with Applications, Vol. 71, Issue C: 469–
478.
Lebkowski, A. 2018. Design of an Autonomous Transport
System for Coastal Areas. TransNav, the International
Journal on Marine Navigation and Safety at Sea, Vol. 12,
No. 1: 117-124.
Lisowski, J. 2014. Optimization-supported decision-making
in the marine game environment. Solid State
Phenomena, Vol. 210: 215-222.
Lisowski, J. 2016. Analysis of methods of determining the
safe ship trajectory. TransNav, the International Journal
of Marine Navigation and Safety of Sea Transportation,
Vol. 10, No. 2:223-228.
Mesterton-Gibbons, M. 2001. An introduction to game
theoretic modeling. Providence: American Mathematical
Society.
Millington, I. & Funge, J. 2009. Artificial intelligence for
games. Amsterdam-Tokyo: Elsevier.
Miloh, T. 1974. Determination of critical manoeuvres for
collision avoidance using the theory of differential
games. Hamburg: Inst. Fur Schiffbau.
Modarres, M. 2006. Risk analysis in engineering. Boca
Raton: Taylor & Francis Group.
Nisan, N., Roughgarden, T., Tardos, E., Vazirani, V.V. 2007.
Algorithmic game theory. New York: Cambridge
University Press.
Olsder, G.J. & Walter, J.L. 1977. A differential game
approach to collision avoidance of ships. Proc. of the 8
th
IFIP Symp. on Optimization Techniques, Novosibirsk,
264-271.
Osborne, M.J. 2004. An introduction to game theory. New
York: Oxford University Press.
Perez, T. 2005. Ship motion control. London: Springer.
Polak, E., Lee. S., Bustany, I., Madhan, A. 2016. Method of
approximations and adaptive approximations for a class
of matrix games. Journal of Optimization Theory and
Applications, Vol. 170, No. 3: 876-899.
Straffin, P.D. 2001. Game theory and strategy. Warszawa:
Scholar.
Szlapczynski, R. & Szlapczynska, J. 2016. An analysis of
domain-based ship collision risk parameters, Ocean
Eng., Vol. 126: 47-56,
Tomera, M. 2012. Nonlinear Observers Design for
Multivariable Ship Motion Control, Polish Maritime
Research, Vol. 19, Special Issue 1: 50-56
Xu, Q. & Wang, N. 2014. A survey on ship collision risk
evaluation. Promet – Traffic & Transportation, Vol. 26,
No. 6: 475-486.
Wells, D. 2013. Games and mathematics. Cambridge:
Cambridge University Press.